You are on page 1of 26

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/271434650

Aircraft Calibration for Application of Solid Baits for Pest Eradication Operations

Conference Paper · January 2011


DOI: 10.13031/2013.37304

CITATIONS READS
0 105

3 authors:

Brian Richardson Mark Owen Kimberley


Scion Envirostat
98 PUBLICATIONS   1,615 CITATIONS    215 PUBLICATIONS   4,236 CITATIONS   

SEE PROFILE SEE PROFILE

Stefan Gous
Scion
42 PUBLICATIONS   189 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Global Planted Forest Trends View project

Trade and Invasive Forest Pests: A case study of ISPM15 View project

All content following this page was uploaded by Mark Owen Kimberley on 09 February 2015.

The user has requested enhancement of the downloaded file.


1 AIRCRAFT CALIBRATION FOR PEST ERADICATION OPERATIONS
2 USING PESTICIDES FORMULATED AS SOLID BAITS

3 B. Richardson, M.O. Kimberley and S.F Gous

4 Abstract. The objective of any aerial (or ground-based) pesticide application is to achieve a biological objective
5 while minimizing costs and environmental and social impacts. A requirement for achieving this objective is to
6 spread the material at a preset rate and at an acceptable uniformity, whilst minimizing the total aircraft travel
7 distance and operation time (which equates to cost). The aircraft calibration procedure is an important step in
8 developing operational parameters to meet these criteria. Many factors influence variability of pesticide deposition
9 during a calibration procedure. However, the calibration process is actually designed to minimize much of this
10 variation. Therefore, pesticide deposit variation during an operation is likely to be significantly higher than
11 expected from the calibration process. The consequence of variable deposition is overdosing and underdosing.
12 Underdosing is a particular concern for a pest eradication operation where survivors could re-establish the
13 population. A new modelling approach proposed for aircraft calibration more accurately quantifies the level of
14 deposit variability likely to result from different operational specifications. Using an operational mosquito
15 eradication programme as a case study, the new analysis highlighted that for this scenario the proportion of the
16 target area underdosed was not extremely sensitive to either size of sampling area or lane separation. Nevertheless,
17 reducing lane separation from the current specification of 20 m to 18 m reduced the predicted underdosed area by
18 about 2-3% depending on the target area size (sample area). Without further analysis, the trade-off between a 2-3%
19 reduction in underdosing (i.e. the probability of eradication) versus an increased operational cost from increased
20 flying time, is a subjective decision. A significant result was the very high sensitivity of the underdosed area to the
21 standard deviation of wind speed. This result means there is a large advantage of maintaining a low wind speed
22 during the application.

23 Keywords. Aerial application, pest eradication, solid material, calibration.

24 INTRODUCTION
25 Biosecurity New Zealand (BNZ) is the branch of the Ministry of Agriculture and Forestry responsible for the
26 New Zealand biosecurity system. Its role includes protecting New Zealand from the introduction of new pests and
27 diseases that could threaten the environment, productive sectors or the health of New Zealanders. Over recent years
28 there have been a number of eradication campaigns undertaken following the detection of unwanted pests. Two
29 recent examples are programs to eradicate the red imported fire ant (RIFA; Solenopsis invicta) and the Southern salt-
30 marsh mosquito (Ochlerotatus camptorhyncus). In both of these cases the eradication effort included the aerial
31 application of pesticides formulated as solid material. S-methoprene, an insect growth regulator formulated as a

1
32 solid, dense granule was applied to water to stop mosquito pupae from hatching into adults. Either pyriproxyfen or
33 methoprene, formulated as a corn-based matrix with an irregular bread-crumb consistency, was applied against
34 RIFA.
35 The objective of any aerial (or ground-based) pesticide application is to achieve a biological objective while
36 minimizing costs and environmental and social impacts. A requirement for achieving this objective when applying
37 pesticides formulated as solid material (or liquids for that matter) is to spread the material at a preset rate and at an
38 acceptable uniformity, whilst minimizing the total travel distance and operation time (which equates to cost) (Grift
39 2000). Aircraft calibration is an important step in developing operational parameters to meet these criteria.
40 According to ASABE Standard S386.2 (ASAE, 1999), calibration of aerial fertilizer equipment is typically
41 undertaken by releasing material from over collectors that are aligned perpendicularly to the aircraft travel path. The
42 amount of material deposited in each collector defines the shape of the swath pattern. Computer software can then
43 be used to overlap the swath pattern with itself, the distance between overlaps being the input lane separation
44 (distance between flight lines). The deposition profile across the spray block is calculated by summing overlapped
45 deposit values. The mean of these values represent the pesticide application rate and the uniformity of the
46 application (or variability) can be calculated as the coefficient of variation (CV) which is defined as:
47
48 CV (%) = standard deviation in deposition x 100 / mean deposition.
49
50 If this procedure is repeated for a range of lane separation values, the relationship between lane separation and
51 CV can be plotted, and there are software products available to automate this analysis (e.g. Joshi et al., 2006; Parish,
52 1987). To maximize aircraft productivity (area sprayed per hour) it is desirable to maximize lane separation as long
53 as overall deposit variability is acceptable. A “robust” pattern has acceptable uniformity for a large range of swath
54 widths (there are minimal changes in CV with flight path errors) (Grift 2000; Grift et al., 2000).
55 A key issue when defining an acceptable CV is to understand the biological consequence of pesticide deposition
56 variability (Richardson et al., 2004). Various rules of thumb have been proposed for different types of application
57 ranging from 15% (Gardisser, 1993) to 30% (Parkin and Wyatt, 1982; Spillman, 1979). Understanding this
58 relationship requires knowledge of the dose-response between the target organism and the applied pesticide and also
59 the distribution of the pesticide deposits (doses) (Richardson et al, 2004).
60 Many factors influence variability of pesticide deposition during calibration procedures or an actual operation.
61 For aerial application of solids, these factors include segregation of material in the aircraft hopper (due to vibrations)
62 leading to a change in released particle size during application (Grift et al 2000); the amount of material in the
63 hopper and the characteristics of the formulation may influence flow rate (Jones et al. 2008; Parish 1999); changes
64 in aircraft speed and height; tracking errors (i.e. not following the specified flight path); variability in wind speed
65 and direction (influences particle trajectory and aircraft flying parameters); and swinging motion of under-slung
66 hoppers (often used with helicopters). Additional experimental variation during a calibration procedure can be
67 caused by solid particles either jumping into or out of the collectors (Parish 1991; Whitney et al 1987).

2
68 A typical aircraft calibration process is actually designed to minimize much of this variation by flying into a light
69 headwind in ideal flying conditions. In some cases, multiple passes over a single line of collectors are undertaken to
70 average out minor run-to-run variations and to provide larger samples to weigh (Parish 1999). Richardson et al.
71 (2004) showed that normal procedures for defining lane separation to match a target CV, will often lead to actual
72 levels of CV that exceed this target value. They also highlighted the importance of knowing the distribution of
73 spray deposits before calculating biological consequences of deposit variation.
74 This article presents results from an operational calibration exercise carried out to define operational parameters
75 for eradication of salt marsh mosquito using a pesticide formulated as a granule and applied using a helicopter. As
76 an operational programme and with very limited funding there were significant constraints on selection of treatments
77 and replication. Nevertheless, the operation provided data that were suitable for evaluating alternative protocols for
78 aircraft calibration.
79

80 METHODS
81 TRIAL LOCATION AND DEPOSIT ASSESSMENT

82 Tests to measure the aircraft swath pattern from a range of treatments were carried out on April 24 2008 at Bridge
83 Pa Aerodrome, Hastings, New Zealand, where the ground surface was flat, short-cropped grass. The approach used
84 to measure the swath pattern was to fly the aircraft over lines of collectors oriented at 90° to the flight line. Bait
85 material was released during the flight and deposited in the collectors. After each treatment, the material deposited
86 in each bucket was placed into small plastic sample containers. The weight of deposited material in each container
87 was measured in the Scion laboratory in Rotorua using a balance (Denver Instrument SI-603) with a precision of 3
88 decimal places.
89 In these trials, a 125 m flight line (95.6°) was marked approximately parallel to the expected wind direction with
90 the flight direction oriented into the wind (Figure 1). In reality the orientation relative to the wind was approximate
91 due to the fluctuating wind direction experienced throughout the trial period.
92 Electric fence standards with attached high-visibility vests marked each end of the flight line. Prior to each
93 treatment, the start and end points of the flight line were logged in the aircraft GPS system (Trimble, Trimflight)
94 with differential correction. They were also recorded using a hand held GPS unit (Garmin GP60).
95 Three 50 m long sampling transects were established at right angles to the direction of flight (Figure 1). The
96 central transect (line B) was located in the middle of the flight line. Transects A and C were 10 m upwind and
97 downwind, respectively, of transect B.
98 Bait collectors were positioned at 2 m intervals along each transect (26 sampling stations per transect). Each
99 sampling station consisted of two buckets but each pair of buckets was treated as one sample. Each bucket had a top
100 diameter of 0.25 m, giving a total sampling area at each station of 0.098 m2. The reason for having two buckets at
101 each sampling station was to ensure that a large enough area was sampled to allow the deposited spray material to be
102 accurately weighed.

3
103 To assess the importance of sampling scale when assessing deposit variation additional collectors (pairs of
104 buckets) were added as follows: (i) a continuous 6 m line of bucket collectors was established along transect B,
105 centered on the intersection with the flight line (i.e. 3 m each side of the centerline); (ii) a similar 6 m line of buckets
106 (no gaps between adjacent buckets) was established along the flight line and centered on the flight line intersection
107 with transect B. These data allowed assessment of deposit variation along the flight line; generally measurements are
108 only taken perpendicular to the flight lines.
109
Flight line 125 m

Expected
wind direction

C: 72.5 m
10 m
125 m B: 62.5 m
10 m
A: 52.5 m

Meteorology:
•Wind speed
•Wind direction
•Temperature
0m •Humidity
•Radiation

110 50 m

111
112 Figure 1: Calibration trial layout.
113

114 APPLICATION METHODS AND TREATMENTS

115 All applications were made using a Bell Jet Ranger 206B III (Helicopters Hawke‟s Bay) fitted with an under-
116 slung bucket applicator for carrying and distributing the mosquito bait. The bucket is a custom made collapsible
117 funnel shaped spinning disk bucket. The bucket is 95 cm deep and has a 90 cm top diameter. At the base, bait is
118 released through an orifice onto a 40 cm diameter spinning disk, powered by a petrol engine. The two disk plates
119 are separated by six, 40 mm high fins. The orifice is opened and closed by a pneumatic, piston driven, sliding plate,
120 controlled by the pilot. The leash attaching the hopper to the helicopter was approximately 7 m long (distance from
121 belly to bucket disc).
122 The treatment variables were specific combinations (Table 1) of release height (25, 30 or 35 m), flying speed (30,
123 39, or 49 knots) and flow rate (23.5 or 25 mm orifice). In total 17 application were made with the treatment of most
124 interest (the standard treatment) being replicated 8 times and other treatments replicated two to three times. For each
125 application, the pilot was requested to record the following information on the aircraft global positioning system:

4
126 flight line point location while spreading (easting/northing) noting points where spreading has started and stopped;
127 date; time; height above sea level; ground speed.
128 All applications were made using methoprene formulated as a sold granule. Methoprene is an insect juvenile
129 hormone analog that acts as a growth regulator and prevents the development of adult mosquitoes. The particle size
130 distribution of the bait is summarized in Figure 2, with the range in diameters from 0.5 – 2.5 mm.
131

Methoprene particle distribution

1100
1087
1000
900
Number of particles

800
700
600
500
488
400
300
200
100 0 60 39 0 0
0
0-500

501-1000

1001-1500

1501-2000

2001-2500

2501-3000

3001-3500
Size Class
132
133 Figure 2: Particle diameter (μm) distribution of the methoprene bait.
134
135 Ideally, the flow rates actually achieved during the test applications would have been recorded in real time.
136 Although such technology has been devised (Grift 2001; Grift et al., 2001), there was no facility to do these
137 measurements in the current trials. As an alternative, the flow rate of the bait was determined prior to the
138 applications by directly measuring the amount of material released from the hopper over a timed period. To achieve
139 these measurements, the hopper was hung from a fork lift truck at about 0.5 m above the ground. A purpose-made
140 “sock” was fitted around the hopper circumference. After starting the spinner motor, the bucket gate was opened and
141 the bait was released for either 30 or 45 seconds. The released bait was then weighed and the flow rate calculated.
142 Flow rate measurements were measured using both a 23.5 (three replicates) and a 25 mm orifice plate (four
143 replicates).
144
145 Table 1: Treatment combinations tested for swath pattern analysis
Test Flying speed Height Orifice Replications
No. (knots) (m) (mm)
1 39 30 23.5
3
4 39 30 25.0
8

5
12 39 35 25.0
2
14 30 30 25.0
2
16 49 30 25.0
2
146
147
148 A Monitor Sensors (Monitor Sensors, Aust. PTY LTD), meteorological station was set up in the vicinity of the
149 flight line (Figure 1). Measurements taken at 1.5 m above the ground were wind speed (km/hr) (anemometer model
150 AN2), wind direction (°) (model WD2 wind direction sensor), temperature (°C) (TA1 ambient temperature sensor),
151 and relative humidity (%) (model HU1 humidity sensor). The sampling time for measurements was 10 seconds and
152 the clock was synchronized with the aircraft GPS system.
153
154

155 ANALYSIS

156 Percentage recovery of methoprene bait


157 Methoprene flow rates were calculated as kg/min for each replicate and an overall mean was calculated for each
158 treatment. Using this information, the theoretical or expected deposition values also were calculated. Actual
159 deposition of bait material was compared with expected deposition to give a percentage recovery.
160
161 Conventional swath pattern analysis
162 For each application and transect a standard analysis was undertaken to determine an appropriate lane separation.
163 However, the relationship between lane separation and coefficient of variation in deposition (CV) was calculated in
164 two ways. Firstly, for each spray run, the individual swath patterns for each transect were averaged and further
165 analyses were undertaken using this averaged swath pattern. This swath pattern was overlapped with itself with a
166 defined spacing between the centers of each swath (i.e. the lane separation). Total deposition was summed for the
167 overlapped patterns, and the CV was calculated by expressing the standard deviation of deposition as a percentage
168 of the mean. This exercise was repeated for a range of lane separation values and the relationship between lane
169 separation and CV was plotted. Lane separation values producing CVs of 20, 30, and 40% were recorded.
170 For the second method, the relationship between lane separation and CV was calculated as described above for
171 each individual transect (as opposed to the mean swath pattern averaged across all three transects). Once the lane
172 separation values corresponding to CVs of 20, 30, and 40% were calculated for each transect, they were averaged to
173 give one value of lane separation for each level of CV.
174
175 Modelling operational variability
176 Arguably, the conventional swath pattern analysis does not provide useful information on the likely distribution
177 of bait deposited during an operation. Without information on the actual bait distribution (i.e. variability in dosage)

6
178 it is not possible to make the link between deposition and likely biological responses, and consequently the
179 definition of an acceptable level of deposit variability (Richardson et al., 2004). Another issue not dealt with using a
180 conventional analysis is defining the appropriate scale at which measurements of deposit variability are made. In
181 theory, using larger sample areas should reduce overall deposit variability. The scale of relevance to the mosquito
182 eradication is variable. In some cases mosquito larvae could be residing in large water bodies and deposit variability
183 may not be a significant issue due to diffusion of the chemical (redistribution) within the water. At the other end of
184 the spectrum, areas as small as water-filled cattle foot prints could also provide a suitable habitat for mosquito
185 larvae.
186 A modelling approach was undertaken to evaluate the importance of these issues. By fitting a general model to
187 individual swath patterns and relating the parameters in the distribution to operational variables, a simulation model
188 was derived. The simulation model enabled a more realistic evaluation of alternative bait application strategies.
189

190 RESULTS AND DISCUSSION


191 AIRCRAFT GPS
192 GPS flight line (point) data were displayed to calculate the track error (offset from the nominal flight line). The
193 flying accuracy was very high with an average cross track error of 0.14 m and a standard deviation of only 0.83 m
194 (Table 2).
195
196 Table 2: GPS output data for each of the 17 calibration tests.

Test Orifice Direction Offset Ground speed Release


number Time dia. (mm) (m) (km/hr) height (m)
1 08:53:25 23.5 E to W 1.5 67.9 41.5
2 09:31:35 23.5 E to W 0.1 72.5 46.5
3 10:02:36 23.5 E to W 1.4 73.1 47.2
4 11:14:14 25 E to W -1.4 70.9 45.7
5 11:47:02 25 E to W 0 71.7 46.6
6 12:10:05 25 E to W 0.3 73.3 43.9
7 12:30:50 25 W to E -0.9 66.3 41.1
8 12:50:59 25 W to E 0 68.3 42.5
9 13:10:33 25 W to E -0.2 67.9 38.3
10 13:28:45 25 W to E 0.5 63.4 39.0
11 13:48:25 25 W to E -0.8 70.4 38.3
12 14:06:09 25 W to E 1.1 68.3 52.3
13 14:25:31 25 W to E 0.3 69.7 51.3
14 14:46:52 25 W to E -0.6 49.0 40.3
15 15:04:44 25 W to E -0.2 55.6 41.2

7
16 15:23:15 25 W to E 0 88.2 47.7
17 15:38:04 25 W to E 1.3 85.4 47.3
Mean 0.14 70 44
1
SD 0.83 9.1 4.3
1
197 Standard deviation
198

199 METEOROLOGY
200 Conditions were extremely favorable throughout the trial (Table 3). Winds were light ranging from 1.4 to 9.0
201 km/hr (mean 2.2 km/hr). Wind direction fluctuated throughout the course of the day. With the flight line oriented at
202 about 96°, the wind direction ranged from a headwind to a crosswind at different times of the day. Temperatures
203 ranged from 14.9 to 19.9°C, and relative humidity from 82 to 53%.
204
205 Table 3: Meteorological conditions averaged from 1 minute before to 1 minute after each application (10

206 second measurement interval).

207

Test Time Wind Relative Temp. Wind Crosswind


number speed humidity (°C) direction speed
(km/hr) (%) (°) (km/hr)
1 08:53:25 4.67 82.35 12.52 273 0.25
2 09:31:35 2.66 77.11 14.91 245 1.36
3 10:02:36 3.97 73.04 15.89 224 3.10
4 11:14:14 1.94 65.81 17.73 209 1.79
5 11:47:02 1.41 64.50 17.88 162 1.29
6 12:10:05 3.62 60.90 18.94 119 1.44
7 12:30:50 6.65 60.32 19.02 77 -2.13
8 12:50:59 6.81 57.59 19.35 48 -5.03
9 13:10:33 4.83 57.92 19.12 67 -2.35
10 13:28:45 4.63 56.74 19.33 144 3.46
11 13:48:25 2.15 52.69 19.89 205 2.03
12 14:06:09 4.31 53.66 19.92 269 0.49
13 14:25:31 6.42 55.27 19.67 106 1.17
14 14:46:52 6.84 59.86 19.56 74 -2.51
15 15:04:44 5.33 60.59 19.47 74 -2.00
16 15:23:15 8.09 63.56 19.08 65 -4.15
17 15:38:04 8.95 65.68 18.52 59 -5.36

8
208
209

210 PERCENTAGE RECOVERY OF METHOPRENE BAIT


211 Methoprene flow rates were very consistent. Mean (and standard deviation) flow rates were 11.685 (0.116)
212 kg/min using a 23.5 mm orifice plate in the hopper and 13.461 (0.113) kg/min with the 25 mm orifice plate. Based
213 on these measured flow rates and actual flying speed, a theoretical application rate was calculated for each test. The
214 actual application rate was estimated by measuring the total deposition in all of the sample buckets (assuming
215 deposition did not extend beyond the transect width) and adjusting for the proportion of the total transect area
216 sampled by the buckets. Using this approach, the total amount of material actually captured was expressed as a
217 proportion of the expected amount based on the hopper output to give a percentage recovery (Table 4).
218
219 Table 4: Percentage of applied material recovered averaged across all three transects.

220
Test Flying Release Orifice Recovery
number speed height diameter
(knots) (m) (mm) (%)
1 39 30 23.5 67.5
2 39 30 23.5 61.4
3 39 30 23.5 58.7
4 39 30 25.0 76.7
5 39 30 25.0 60.8
6 39 30 25.0 72.7
7 39 30 25.0 96.3
8 39 30 25.0 71.3
9 39 25 25.0 82.3
10 39 25 25.0 78.9
11 39 25 25.0 75.4
12 39 35 25.0 79.4
13 39 35 25.0 68.9
14 30 30 25.0 68.1
15 30 30 25.0 78.0
16 49 30 25.0 66.0
17 49 30 25.0 79.1
221
222

223 The overall mean recovery was 73% leaving 27% of the material unaccounted for. There are two possible
224 explanations for this result:

9
225 1. The methoprene flow rate was lower during the application than in the static flow rate tests. This explanation is
226 unlikely given the free-flowing nature of the product and the low variance in the measured flow rates.
227 2. The most likely explanation is that some of the methoprene granules bounced out after landing in the buckets.
228 This is a well-known phenomenon (Parish, 1991; 2000). A simple test of this hypothesis was undertaken in the
229 laboratory. These tests demonstrated that it is highly likely that some granules were indeed bouncing out of the
230 buckets. This conclusion indicates that future work with these granules should use a modified sampling system
231 to reduce loss of granules and to maximize overall recovery. Since the purpose of the work described here was
232 to use the operational data to review the underlying theory of the calibration process, this inconsistency between
233 measured and expected deposition was not viewed as critical.
234
235 CONVENTIONAL SWATH PATTERN ANALYSIS
236 As described in the methods, a conventional swath pattern analysis was undertaken using two averaging methods.
237 The first method was based on averaging the three transects for each test and then developing a relationship between
238 lane separation and CV using this average pattern (“combined data” in Table 5). For the second method, the
239 relationship between lane separation and CV was developed for each individual transect within a test. The lane
240 separation values for a given level of CV were averaged subsequently (“individual data” in Table 5).
241
242 Table 5: Calculated lane separation values for each treatment

Test Speed Ht Orifice Combined dataa: Lane Individual datab: Mean


number (knots) (m) (mm) separation for CVs lane separation for CVs
20% 30% 40% 20% 30% 40%
1 39 30 23.5 17.8 19.8 21.8 11.3 15.3 18.8
2 39 30 23.5 18.5 20.3 22.0 9.7 12.0 18.5
3 39 30 23.5 13.3 19.0 22.0 7.3 11.3 17.4
4 39 30 25.0 18.5 20.8 22.8 12.4 18.3 20.6
5 39 30 25.0 20.0 22.5 24.5 14.5 17.3 19.7
6 39 30 25.0 19.3 22.3 24.0 14.3 17.6 23.1
7 39 30 25.0 19.3 21.5 23.5 15.4 20.2 22.6
8 39 30 25.0 18.5 21.0 23.0 13.4 17.5 20.5
9 39 25 25.0 18.3 20.8 23.0 14.7 19.2 21.3
10 39 25 25.0 20.0 21.8 23.5 13.9 20.7 22.4
11 39 25 25.0 15.5 18.5 20.5 13.8 17.2 19.7
12 39 35 25.0 21.3 23.5 25.5 11.5 19.6 22.9
13 39 35 25.0 18.8 21.0 23.0 11.3 15.8 19.8
14 30 30 25.0 19.8 22.3 24.3 15.5 20.7 22.9
15 30 30 25.0 9.0 19.3 21.3 10.2 15.8 18.6
16 49 30 25.0 13.0 17.3 19.3 9.1 12.6 17.2

10
17 49 30 25.0 19.3 21.3 23.3 11.27 13.1 17.7
a
243 The mean swath pattern was created by averaging Transects A, B and C. Lane separation values for each CV
244 were calculated from this average swath pattern.
b
245 Lane separation values for each CV were calculated for each individual transect. The mean lane separation for
246 each CV was based on the average lane separation for each transect.
247

248 As expected, the analysis based on the combined data significantly reduced overall variability by developing a
249 smoothed or averaged swath pattern. Consequently, using the averaged pattern led to a larger lane separation for a
250 given acceptable level of deposit variability (CV). The consequence of using this lane separation in the field,
251 however, would be larger actual variability than expected. The next step in the analysis was to more clearly
252 understand the probability of achieving the target application rate at a scale that is meaningful for the mosquito
253 eradication programme.
254

255 MODELLING OPERATIONAL VARIABILITY


256 As described previously, each collector consisted of a pair of 0.25 m diameter buckets with the total area of 0.098
257 m2. Contents of each collector were weighed and converted into application rates, (in kg/ha) after each run.
258 The distribution of material (in kg/ha) across a swath was well approximated by a Gaussian distribution,
259

260 [1]   
Weight  W S 2 exp  d  M  2S 2
2

261
262 where Weight (kg/ha) is the weight at distance d (m) along the swath, with d centered under the flight path (i.e., d
263 is negative to the left and positive to the right of the flight path). The parameters of the distribution are W (kg/m), the
264 weight of material deposited from the aircraft per metre of flight path, S (m) which controls the width of spread
265 across the swath, and M (m) which is the location of the midpoint of the distribution relative to the flight path.
266
267 The distribution parameters were estimated for each run by nonlinear regression using the SAS procedure NLIN.
268 Parameter estimates are given in the Appendix. Figure 3 shows the Gaussian distributions plotted against actual
269 measurements for two representative runs. In most cases, the Gaussian distribution closely approximated the actual
270 distribution with the regression R2 being greater than 75% for all runs (Appendix).

11
12

10
Weight (kg/ha)

0
-25 -15 -5 5 15 25
Distance from nominal centre of flight line (m)
271
272 Figure 3: Measured and predicted (using Model [1]) deposition at right angles to the flight line for two
273 representative runs. Results are shown for a low crosswind speed (Run 1, measurements shown by diamonds,
274 predictions by solid line) and a higher crosswind speed (Run 17, measurements shown by triangles,
275 predictions by dashed line).
276

277 Relationships between distributional parameters and flight variables


278 The distribution midpoint parameter M was closely related to crosswind speed (Figure 4), and to a lesser extent,
279 to the measured offset, and to the direction of the flight (E to W versus W to E). These relationships were
280 demonstrated by a regression model (Table 6) from which the following can be concluded: (i) for every 1 km/hr
281 increase in crosswind speed, the midpoint of the distribution moved about 0.9 m; (ii) the distribution had a
282 consistent bias of about 0.9 m to the left of the flightline in the direction of flight – this can be inferred from the fact
283 that when the aircraft switched direction, the intercept moved 1.8 m. This slight bias is possibly due to the rotation
284 of the spinner in the hopper. The fact that such a slight bias is detectable from these results reflects the high quality
285 of the data collected; and (iii) the measured offset from the centerline of the trial was detectable in the fitted
286 distribution with an equivalent shift in M (i.e., the coefficient for offset in the model was statistically significant and
287 close to -1).
288

12
6
4

0
M (m)

-2

-4

-6
-8

-10
-12
-6 -4 -2 0 2 4
Crosswind speed (km/hr)
289
290 Figure 4: Relationship between M and crosswind speed.
291
292
293 Table 6. Regression model for predicting distributional midpoint parameter from run variables. R 2 = 0.92;
294 Root MSE = 1.08
Parameter Estimate F-ratio P-value
Intercept (E to W) -0.6
Intercept (W to E) -2.4 8.12 0.014
Crosswind speed 0.941 67.85 <.0001
Offset -1.2 13.72 0.0027
295
296 The parameter S, which controls the width of the distribution across the swath, was not significantly related to
297 any of the run variables (i.e., flight speed, wind speed, height, or plate size). This is a positive result for the
298 calibration process because it implies that any small changes needed in flying speed or release height will not
299 influence the width of the swath pattern.
300 As expected, the parameter W was directly proportional to the inverse of flight speed and was 26% lower for the
301 23.5 mm plate than the 25 mm plate. This demonstrates that the deposition results were sensitive to changes in
302 output over the sample lines due to either flying speed or flow rate through the plate orifice.
303

304 Small-scale and large-scale variability


305 The above analysis shows that it is possible to model the swath pattern for the described conditions using a
306 Gaussian distribution, with distributional parameters controlled by flight speed and direction, orifice plate size, flight
307 offset, and wind speed and direction. However, to accurately simulate deposit distributions, it is also necessary to

13
308 understand the small-scale and large-scale variability of deposits about the mean distribution. In the following
309 analysis, the experimental unit consists of a collector of 0.098 m2 area as used in the trial.
310 Firstly, we tested the data for spatial autocorrelation. If spatial autocorrelation is present, the variation between
311 widely spaced collectors should be greater than the variation between closely spaced buckets. To test this, the pooled
312 variance of residuals (actual – Gaussian predicted deposition) of pairs of collectors at different spacings both along
313 and across the flight line was calculated.
314 Along the flight line, a spacing of 20 m was achieved using central collectors in the 1st and 3rd transects of each
315 run. A 10 m spacing was achieved using the 1st and 2nd, and the 2nd and 3rd transects. Closer spacings were achieved
316 using the 6 m long array of closely spaced collectors placed along the flight line. Cross-flight line spacings of 6 m
317 and less were achieved using the closely spaced array of collectors placed at the centre of Transect B. In both
318 directions, the closest spacing of 0.25 m was achieved using pairs of adjacent collectors.
319 The pooled variance of deposits calculated using pairs of collectors at different spacings is shown in Figure 5.
320 There was no trend of the variance decreasing at closer spacings, indicating that there was little spatial
321 autocorrelation. Also, the variance was similar both along and across the flightline.
322

2.5

2.0
Variance of residuals

1.5

1.0

0.5
Along Flightline
Across Flightline
0.0
0 5 10 15 20
Spacing (m)
323
324 Figure 5: Variance of pairs of residuals at different spacings along and across the flightline.
325 It can therefore be concluded that apart from the systematic cross-sectional swath pattern described above,
326 deposition was completely random, with no large-scale or small-scale clumping of material, at least up to a distance
327 of 20 m.
328 In addition to testing for autocorrelation, it was also necessary to determine whether the variance varied with the
329 mean deposition. If deposition is randomly distributed, weights in collectors should follow a Poisson distribution. In
330 Poisson-type distributions, the variance is proportional to the mean, and it was expected that this might be the case
331 in this trial.

14
332 To test this hypothesis, collectors were classified on the basis of their predicted deposition using classes of <0.25,
333 0.25-0.5, 0.5-1, 1-2, 2-3, 4-5, 5-6, and >6 kg/ha. The mean predicted deposition and variance of the residuals (actual
334 – predicted) was calculated for each class. The variance was found to be proportional to the mean predicted
335 deposition (Figure 6). The following no-intercept regression equation for predicting variance from predicted mean
336 deposition was fitted: Variance = 0.447 × Mean; R2 = 0.97.

3.5
Variance of residuals

2.5

1.5

0.5

0
0 1 2 3 4 5 6 7 8
Predicted deposition (kg/ha)
337
338 Figure 6: Variance of residuals versus predicted deposition.
339
340

341 SIMULATION
342 Using the results from the above analysis, a simulation system was developed using the programming language
343 SAS (Version 9.1) to represent the deposition of material for a typical operation. This simulation system was then
344 used to predict various properties of the deposit distribution. In particular, it was used to predict the distribution of
345 active material deposited in theoretical pools of water of different sizes (i.e. mosquito habitat). From this simulation,
346 it was possible to predict the percentage of pools of a given size receiving lethal doses of active material.
347 The programme simulated a cross-section of an operational area. The simulation predicted deposits in squares
348 with 0.33 m sides. These are of similar size to the collectors used in the trial (although the trial used pairs of circular
349 buckets as collectors, the squares used in the simulation can be considered to closely approximate these.)
350 To achieve a stable prediction, 2000 flight paths were included in the simulation. Sufficient contiguous samples
351 were simulated in the flightline direction to accommodate simulated pools of water of different sizes. Pools
352 consisted of square groups of samples. For example, single samples were used to represent 0.1 m2 pools, 2×2 sample
353 groups to represent 0.4 m2 pools, and 3×3 sample groups to represent 1 m2 pools.
354 The simulation operated as follows. Gaussian distributional parameters are generated for each flight line using
355 results from the above analysis of the trial.

15
356 The parameter M was generated using the equation:
357
358 [2] M = Direction_Effect + Offset + 0.94 × Crosswind_Speed + E
359
360 where, based on the trial data,
361  Direction_Effect alternated between 0.9 m and -0.9 m for each flight line (representing the aircraft flying
362 back and forth).
363  Offset, which represents cross track error, was generated from a random normal distribution with mean 0 and
364 standard deviation 0.83 m.
365  Crosswind_Speed was generated from a random normal distribution with mean 0 and standard deviation 2.8
366 km/hr, the latter being the standard deviation of the measured crosswind speed (Table 3).
367  E was generated from a random normal distribution with mean 0 and standard deviation 1.08 m
368
369 The parameter S was generated from a random normal distribution with mean 6.08 m and standard deviation 0.53
370 m. The parameter W which represents the weight of material deposited from the aircraft per metre of flight path, can
371 be calculated for a given lane spacing and application rate as W = L × A (where L = lane separation and A =
372 application rate). Rather than using the mean value of W obtained in the trial which may be too low, possibly
373 because of bounce-loss of material from the collectors, we assumed that the operational mean application rate can be
374 accurately achieved by adjusting flight speed or orifice size. The operational application rate used was 6 kg/ha and
375 the flight line spacing was 20 m implying a mean W of 120. The standard deviation of W in trial runs which used the
376 standard plate and air speed (runs 4-13) was 17.3% of the mean. Therefore, in the simulation, W was generated from
377 a random normal distribution with mean 120 km/m and standard deviation 20.8.
378 The values of M, W and S generated for each flight pass were used to predict the expected deposition at each
379 sample location in the simulated operation. The local deposition for each 0.1 m2 sample was generated using a
380 normal distribution with Variance = 0.447 × Expected Depostition.
381 It is possible to validate the simulation model by comparing its results with those obtained using a conventional
382 swath pattern analysis. To do this, a single swath pattern was simulated and overlapped, and the CV of the
383 overlapped simulated deposition was calculated. This procedure was repeated 1000 times to obtain a stable mean
384 CV. Values were obtained both for simulated individual transects, and for combined data from 3 simulated transects,
385 using lane separations of 12 to 22 m in steps of 2 m. The mean CVs of these simulations agreed closely with those
386 obtained using the conventional swath pattern analysis of measured transects (Figure 7), confirming that the
387 simulation system produces swath patterns similar to those obtained experimentally.
388

16
70

60

50

40
CV (%)

30

20 Sim. full random


Conv. individual
10 Sim. individual
Conv. combined
Sim. combined
0
10 12 14 16 18 20 22 24

Lane spacing (m)


389
390 Figure 7: Comparison of mean CVs for varying lane separations calculated using conventional swath
391 pattern analysis of actual transect data, and simulated data. Values are shown for individual transects, and
392 for combined swaths of 3 combined transects. Also shown are CVs of individual transects predicted using the
393 full simulation system.
394
395 Because the conventional analysis operates by overlapping the same swath pattern, it does not take account of
396 variation in wind speed, cross track error, flow rate, etc. In contrast, the simulation system described above accounts
397 for these additional sources of variation. The CVs were calculated for the full simulation system are therefore much
398 larger than those obtained using the conventional analysis (Figure 7). The simulation system also allows the relative
399 importance of each source of variation to be evaluated. This can be done by running the simulation with a single
400 input or parameter varying randomly while keeping the remaining parameters and inputs fixed. The results of doing
401 this for lane separations of 12 and 18 m are shown in Table 7. This shows the CV for simulated overlapped Gaussian
402 distributions with no error terms, with only local variation in deposition, with local variation plus variable wind
403 speed, etc. Also shown are mean CVs from the conventional swath pattern analysis and for simulated conventional
404 analysis. Note that the simulated conventional analysis is similar to the simulation with „Local‟ variation except that
405 it uses a single overlapped transect rather than separately simulated transects. Table 7 shows that crosswind speed
406 (with standard deviation 2.8 km/hr), has the greatest effect on deposition CV.
407
408 Table 7: Coefficients of variation (%) for lane separations of 12 and 18 m. Values given are the mean of
409 individual transect conventional swath pattern analyses, simulated conventional analysis, simulations
410 showing the effects of various sources of variation in inputs and parameters, and simulation with all sources
411 of variation.

17
Sources of
12 m lane 18 m lane
Method variation included in
separation separation
simulation
Conventional swath pattern analysis 22.0 29.7
Simulation of conventional analysis 23.5 28.7
None 18.0 25.8
Local 26.4 30.6
Local + Crosswind 45.1 51.8
Simulations with inputs and parameters Local + Offset 32.4 37.8
either held constant or allowed to vary Local + E 32.5 38.0
Local + S 32.0 38.1
Local + W 31.6 37.1
All sources 48.7 57.3
412
413 Figures 8-10 are base on an example 300 m cross-section of bait deposition taken from the simulation. Figure 8
414 shows the mean deposition over this section of the simulated operation. In Figures 9 and 10, the simulated
415 deposition in sample units representing pools of water of 0.1 m2 (Figure 9) and 1 m2 (Figure 10) are shown for the
416 same cross-section. The simulated deposition distributions for pools of size varying from 0.1 m2 to 16 m2 for the
417 entire simulation are shown in Figure 11. In Tables 8 and 9, the percentage of samples receiving concentrations
418 below 4 and 2 kg/ha respectively are shown for various simulations. These include the standard simulation with a 20
419 m flight path spacing, and simulations with 18 and 22 m spacings. Also included is a simulation with twice the
420 observed variability in wind speed (std. dev. = 5.8) to demonstrate the effect of increased wind speed variation on
421 deposition.
422
423
424

18
15

10
Weight (kg/ha)

0
100 150 200 250 300 350 400
Distance (m)
425
426 Figure 8: Simulated mean deposition over a 300 metre cross section of the operation.
427
428
15

10
Weight (kg/ha)

0
100 150 200 250 300 350 400
Distance (m)
429
430 Figure 9: Simulated deposition in 0.1 m2 sample units over a 300 metre cross section of the operation.
431
432
433
434
435
436
437
438
439

19
15

10
Weight (kg/ha)

0
100 150 200 250 300 350 400
Distance (m)
440
441 Figure 10: Simulated deposition in 1 m2 sample units over a 300 metre cross section of the operation.
442

18
0.1
16 0.4
14 1
12 4
% of area

16
10
8
6
4
2
0
0
1

11

12

13
0-

1-

2-

3-

4-

5-

6-

7-

8-

-
9-

10

11

12

Application rate (kg/ha)


443
444 Figure 11: Simulated distribution of concentrations in 0.1, 0.4, 1, 4, and 16 m2 pools.
445
446
447

20
448 Table 8: Percentage of samples with less than 4 kg/ha of active ingredient for various scenarios and
449 samples sizes.
Wind speed SD = 2.8 Wind speed SD =5.6
Sample area (m2)
Lane = 18 m Lane = 20 m Lane = 22 m Lane = 20 m
0.1 24.1 26.5 28.9 31.6
0.4 20.0 23.1 26.1 29.3
1 19.2 22.4 25.7 28.9
4 18.7 21.7 25.2 28.6
16 18.0 21.1 24.2 28.1
450
451
452 Table 9: Percentage of samples with less than 2 kg/ha of active ingredient for various scenarios and
453 samples sizes.
Windspeed SD = 2.8 Windspeed SD =5.6
2
Sample area (m ) Lane = Lane =
Lane = 18m Lane = 20m
20m 22m
0.1 6.1 7.9 10.1 14.2
0.4 3.7 5.7 8.1 12.5
1 3.4 5.3 7.5 12.1
4 3.1 5.1 7.1 11.9
16 2.8 4.4 6.5 11.4
454
455
456 Results of the simulation, particularly those highlighted in Tables 8 and 9 illustrate a number of key points:

457  For lane separation values of 18 to 22 m, every 2 m increase in lane separation increases the target area that
458 receives a dose of less than 4 kg/ha by about 2.4 to 3.5%. Reducing the lane separation from the current value
459 of 20 m to 18 m would reduce the underdosed area by about 3 % depending on the target area size (sample
460 area).

461  For lane separation values of 18 to 22 m, every 2 m increase in lane separation increases the target area that
462 receives a dose of less than 2 kg/ha by about 1.6 to 2.4%. Reducing the lane separation from the current value
463 of 20 m to 18 m would reduce the underdosed area by about 2% depending on the target area size (sample area).

464  The sensitivity of underdosed area to sample size (noting that the sample size is equivalent to the area of target
465 water bodies within the application zone) is not extreme.

466  A significant result was the very high sensitivity of the underdosed area to the standard deviation of wind speed.
467 In general, as wind speed increases, the standard deviation of wind speed also increases (Figure 12). This result
468 means there is a large advantage of maintaining a low wind speed during the application.

21
469

Wind speed s. dev. (km/hr) 3.0

2.5 y = 0.2138x
R2 = 0.3294
2.0

1.5

1.0

0.5

0.0
0 2 4 6 8 10
Wind speed (km/hr)
470
471 Figure 12: Simulated distribution of concentrations in 0.1, 0.4, 1, 4, and 16 m2 pools.
472
473 To truly understand the implications for a pest eradication programme of increasing the underdosed areas by
474 around 2-3% would require this work to be integrated with models of population dynamics. In the absence of such a
475 model, the decision on whether an increase in probability of underdosing of this magnitude is significant is a
476 subjective decision balancing the risk of lost aircraft productivity (increased cost) from reducing lane separation (in
477 this case from 20 to 18 m) against the likelihood that some insects may survive the application and re-establish a
478 population.
479

480 CONCLUSIONS

481 The application system for eradication of salt marsh mosquito using methoprene bait produced highly repeatable
482 results. However, during swath pattern measurements, only approximately 73% of applied bait was recovered. It is
483 highly likely that at least part of the discrepancy between the theoretical and actual recovery of applied bait was due
484 to granules bouncing out of the bucket collectors. The only other realistic explanation is that flow rates during
485 calibration are different to those during flying. Despite the reduced overall catch of material, results were extremely
486 consistent and the analysis of variability of bait deposition (the key purpose of this experiment) was not
487 compromised.
488 The conventional approach to swath pattern analysis is not adequate for an eradication programme where it is
489 critical to minimize the probability of significant underdosing leading to unacceptable numbers of surviving insects.
490 When using the conventional approach to swath pattern analysis, care must be taken not to unrealistically smooth
491 the mean swath pattern by averaging results from adjacent sampling transects. Over-averaging results artificially
492 reduces the sensitivity of deposit variation to increased lane separation. The conventional approach to swath pattern

22
493 analysis does not easily enable the effect of flight and weather variables to be factored into the relationship between
494 deposit variability and lane separation
495 A new modelling/simulation approach to swath pattern analysis and calibration is proposed for eradication
496 operations or situations where there is high sensitivity to deposit variation. Using an operational mosquito
497 eradication programme as a case study, the new analysis highlighted that for this scenario the proportion of the
498 target area underdosed was not extremely sensitive to either size of sampling area or lane separation. Nevertheless,
499 reducing lane separation from the current specification of 20 m to 18 m would reduce the underdosed area by about
500 2-3% depending on the target area size (sample area). Without further analysis, the trade-off between a 2-3%
501 reduction in underdosing (i.e. and index of the probability of eradication) versus an increased operational cost from
502 increased flying time, is a subjective decision.
503 A significant result was the very high sensitivity of the underdosed area to the standard deviation of wind speed.
504 This result means there is a large advantage of maintaining a low wind speed during the application.
505

506 ACKNOWLEDGEMENTS
507 Many people contributed to the collection of data presented in this report with funding from the Ministry of
508 Agriculture and Forestry. Members of the AGM Technical Advisory Group contributed many ideas and suggestions
509 during the course of the spray operation. This work was funded by New Zealand's Foundation for Research, Science
510 & Technology through contract CO2X0501, the Better Border Biosecurity (B3) programme (www.b3nz.org).

511

512 REFERENCES

513 ASAE Standards 1999. S386.2 (R2009). Calibration and distribution pattern testing of agricultural aerial
514 application equipment. St. Joseph, Mich.: ASAE.
515 Gardisser, D.R. 1993. Agricultural aircraft setup. Pub. MP351. Little Rock, Ark.: Cooperative Extension
516 Service, University of Arkansas.
517 Grift, T.E. 2000. Spread pattern analysis tool (SPAT): I. Development and theoretical examples. Trans. ASAE
518 43(6): 1341-1350.
519 Grift, T.E. 2001. Mass flow measurement of granular materials in aerial application. Part 1. Simulation and
520 modeling. Trans ASAE 44(1):19-26.
521 Grift, T.E., J.T. Walker, and D.R. Gardisser. 2000. Spread pattern analysis tool (SPAT): II. Examples of
522 aircraft pattern analysis. Trans. ASAE 43(6): 1351-1362.
523 Grift, T.E., J.T. Walker, And Hofstee, J.W. 2001. Mass flow measurement of granular materials in aerial
524 application. Part 2. Experimental model validation. Trans ASAE 44(1): 27-34.
525 Jones, J.R., R.I.Murray, R.I., and I.J. Yule. 2008. Modelling the coarse fraction of solid fertilizer deposition
526 from a fixed-wing aircraft: I. A ballistics model. Trans. ASABE 51(3): 857-872.

23
527 Joshi, M., N. Giannico, and R.L. Parish. 2006. Improved computer program for spreader pattern analysis.
528 Applied Engineering in Agriculture 22(6): 799-800.
529 May, K.R. and Clifford, R. 1967. The impaction of aerosol particles on cylinders, spheres, ribbons and discs.
530 Annals of Occupational Hygiene 10: 83-95.
531 Murray, R.I., and I.J. Yule. 2007. Developing variable rate application technology: economic impact for farm
532 owners and topdressing operators. New Zealand. J. Agric. Research 50: 65-72.
533 Parish, R.L. 1987. A computer program for spreader pattern analysis. Applied Engineering in Agriculture 3(1):
534 14-16.
535 Parish, R.L. 1991. Effect of material bouncing into and out of collection pans on observed spreader
536 distribution pattern. Applied Engineering in Agriculture 7(3): 311-315.
537 Parish, R.L. 1999. The effect of multiple passes on spreader pattern test results. Applied Engineering in
538 Agriculture 15(6): 643-645.
539 Parish, R.L. 2000. Spreader rate determinations from pattern tests compared with rate calibration. Applied
540 Engineering in Agriculture 16(2): 119-120.
541 Parkin, C.S. and Wyatt, J.C. 1982. The determination of flight-lane separations for the aerial application of
542 herbicides. Crop Protection 1: 309-321.
543 Richardson B, Moore JR, Kimberley MO, Haslett M 2003. Monitoring and modelling aerial spray applications
544 for pest eradication operations. ASAE Paper No. 03-1092, American Society of Agricultural Engineers, St
545 Joseph, MI, USA.
546 Richardson, B., Kay, M.K., Kimberley, M.O., Charles, J.G. and Gresham, B.A. 2005. Evaluating the benefits
547 of dose-response bioassays during aerial pest eradication operations. New Zealand Plant Protection 58: 17-23.
548 Richardson, B., Kimberley, M.O. and Schou, W.C. 2004. Defining acceptable levels of herbicide deposit
549 variation from aerial spraying. Applied Engineering in Agriculture 20: 259-267.
550 Spillman, J.J. 1979. The efficiency of aerial spraying. Convention of the Aerial Agricultural Association of
551 Australia, Perth, Australia, 28-31 May, 1979. Paper No. 783.
552 Streibig, J.C., Rudemo, M., Jensen, J.E. 1993. Dose-response curves and statistical models. In: Streibig, J.C.,
553 Kudsk, P. ed. Herbicide Bioassays. CRC Press, Boca Raton, FL. Pp. 29-55.
554 Teske, M.E., H.W. Thistle, and I.J. Grob. 2007. Determination of dry material physical characteristics for use
555 in dispersion modeling. Trans. ASAE 50(4): 1149-1156.
556 Thistle, H., R. Reardon, M. Teske, B. Richardson, G. Cormier, D. Davies, S, Cameron, A. Hewitt, M. LeClerc
557 and A. Karipot. 2005. 'Variability in Spray Application and Utilization of a Time Varying Model'. Annual
558 Review of Agricultural Engineering (1) 187-196.
559 Whitney, R.W., L.O. Roth, and D.K. Kuhlman 1987a. Deposition uniformity of aerially applied granules.
560 Trans ASAE. 30(2): 332-337.
561

24
562 APPENDIX – GAUSSIAN DISTRIBUTION PARAMETERS AND PERCENTAGE VARIANCE EXPLAINED (R2) FOR
563 NONLINEAR REGRESSION MODELS (EQUATION [1]) FITTED FOR EACH RUN

564

Run M S W R2
1 -1.60 5.66 77.3 87.3
2 -0.40 5.60 65.5 81.3
3 1.60 5.72 58.5 78.4
4 3.77 6.49 90.9 82.1
5 -1.10 6.32 73.8 85.0
6 0.69 6.64 82.8 85.7
7 -4.28 6.41 130.8 86.4
8 -6.07 6.43 94.7 79.8
9 -4.09 6.25 104.8 89.2
10 -0.76 6.06 110.2 86.2
11 0.89 5.35 87.9 94.3
12 -3.82 7.19 105.0 75.6
13 -0.76 6.20 82.4 83.1
14 -5.04 6.64 118.6 86.9
15 -2.38 5.47 121.7 83.5
16 -6.87 5.40 63.8 84.6
17 -9.63 5.59 73.6 75.4

Mean -2.34 6.08 90.7


Std
dev 3.42 0.53 21.5
565
566
567

25

View publication stats

You might also like